Self-hosted autonomous coding agent with team-knowledge RAG.
Runs against your local LLM (Ollama / qwen / llama / mistral) or any provider.
No cloud dependency required.
What OKAI is
OKAI is a VS Code extension that gives you an autonomous coding agent inside your IDE. The agent reads your code, plans changes, runs commands, and edits files — all with your approval, every step of the way.
What makes OKAI different from cloud-based assistants:
Self-hosted by default. Connect to your own Ollama instance, or bring keys for any provider. No phone-home, no third-party data sharing.
Team-knowledge RAG via MCP. Connect OKAI to your team's design guides, ADRs, conventions, and any other context your developers rely on — via the Model Context Protocol.
Full transparency. Every file read and every proposed diff is shown before it's applied. Checkpoints let you revert anything.
Your code stays on your machine. Tool execution happens locally — OKAI orchestrates, your environment runs the code.
Getting started
Install the extension from VS Code Marketplace (or from a .vsix build).
Open the OKAI sidebar (activity bar icon).
Configure your model provider:
Local: point Ollama at qwen2.5:14b or any other model
Cloud: bring an API key for Anthropic, OpenAI, Google Gemini, OpenRouter, etc.
Optionally connect your team's MCP servers for RAG, databases, APIs.
npm install
cd webview-ui && npm install && cd ..
npm run protos
npm run compile
Then F5 in VS Code to launch an Extension Development Host with OKAI installed.
Attribution
OKAI is a fork of Cline (Apache 2.0), an open-source coding agent created by Saoud Rizwan and contributors. We're grateful for Cline's work and its open-source license that made this fork possible. See LICENSE and NOTICE for details.
The OKAI fork adds:
Self-hosted-first defaults (telemetry disabled, no cline.bot phone-home)
Tight integration with the OKAI backend's RAG knowledge base, GitHub/GitLab repo support, and findings store (via MCP)
